Radiology Imaging Slot No-Show Prediction

This project predicts patient no-shows for Imaging appointments using machine learning.
It combines the public Kaggle "No-show Appointments" dataset with synthetic healthcare scheduling data to simulate real hospital scheduling operations.

Objectives

• Analyze appointment and patient patterns contributing to no-shows
• Predict which appointments are at high risk of being missed
• Provide actionable insights for schedulers and hospital administrators
• Support integration with an Scheduling slot-management app

Problem Addressed

Missed appointments cost hospitals both time and resources. By predicting potential no-shows, facilities can:

• Reallocate slots efficiently
• Reduce idle machine time
• Improve patient access and care coordination

Dataset

• Source: Kaggle No-show Appointments Dataset
• Synthetic Features Added:
  • Medical_Transport (Yes/No)
  • `Appointment_Day
  • WaitingDays
  • Visit_Count
  • Past_Noshow_count

Exploratory Data Analysis (EDA)

• Distribution of no-shows by age, gender, waiting days, Medical_Transport, Appointment_Day and more
• Chi-square tests for feature significance
• Correlation heatmaps and feature selection

Predictive Modelling

Models tested:

• Logistic Regression
• Random Forest
• XGBoost

Handling Class Imbalance: SMOTE applied to balance show/no-show cases.

Confusion matrices and performance plots available in the /images folder.

Model Comparison (No Show class @ Best thresholds)

Model	Recall	Precision	F1
Logistic Regression	0.65	0.31	0.42
Logistic Regression (Smote)	0.60	0.26	0.37
Random Forest	0.74	0.28	0.41
XGBoost (final)	0.87	0.30	0.44

Confusion Matrix from XGBoost Model

Business Impact

Predictive insights can help clinic/hospitals:

• Identify high-risk patients
• Send targeted reminders
• Improve machine utilization and reduce missed idle slots

Future Scope

• Integrate model output into Power Apps Imaging Scheduling Tool
• Add real-time prediction dashboard in Power BI

Credits and Acknowledgments

• Original dataset by Joni Arroba on Kaggle (https://www.kaggle.com/datasets/joniarroba/noshowappointments)
• Synthetic data creation, analysis, and modelling by Prashasti Hajela
• Tools used: Jupyter Notebook, Anaconda, Python, Pandas, NumPy,Matplotlib, Seaborn, Scikit-learn, XGBoost, Imbalanced-learn

Contact Information

For questions, suggestions, or collaboration:

• Author: Prashasti Hajela
• Email: [email protected]
• LinkedIn: https://www.linkedin.com/in/prashasti-h-73bb4b137/